Track 16 · Advanced Foundations · Lesson 7 of 8
Causal inference: potential outcomes and causal diagrams
Define causal effects with potential outcomes, use simple causal diagrams to decide what to adjust for, and explain what randomisation and regression adjustment can and cannot achieve.
Tutored students scored lower on the final exam than untutored ones. Does tutoring hurt? Or were the students who were offered tutoring already struggling? Answering “what would have happened otherwise?” is the core of causal inference — and no amount of data answers it without a design or assumptions.
Potential outcomes. For each unit (a student, a patient) imagine two outcomes: if the unit receives the treatment, and if it does not. The individual causal effect is . We only ever see one of the two — the other is a counterfactual. This fundamental problem of causal inference is why causal questions need more than data: they need a way to estimate the outcome we did not see.
Included with a Formula Quest membership
It has 8 graded exercises with feedback, hints and worked solutions. After it you will be able to:
- Define individual and average causal effects with potential outcomes
- Explain how selection bias makes naive comparisons misleading and why randomisation fixes it
- Classify variables as confounders, mediators or colliders and choose what to adjust for
- Standardise rates and recognise Simpson's paradox
- State what regression adjustment can and cannot establish